Final Project EDA

Pitch Mix

During my first quarter project I presented the following visualization that shows Jacob deGrom’s pitch mix over the years in the data.

From the plot you can see that Jacob deGrom trusted his 4-seam fastball more than he did in previous years in addition to his slider.

Next I wanted to see how whether or not if the pitch was a strike out or a non-strikeout pitch how that affected deGrom’s pitch choice.

From the above plots you can see that the increase in deGrom’s fastball usage actually most likely came from his increased usage on non-strikeout pitches and deGrom got less strikeouts on his fastball than he did in years past. It also appears deGrom all around used his slider more often.

Result of Pitch

Again in my first quarter project I presented a visualization that showed the proportions of strikes thrown, balls thrown, and balls hit in play. The plot below shows that deGrom threw more strikes last year while throwing less balls and allowing less balls hit in play.

Next I wanted to break this proportions up by pitch, so I changed the denominator to be the amount of times deGrom through that pitch in that season and made the numerator the amount of times that pitch was a strike, ball, or hit in play. The plot below is the result.

It it pretty interesting to note that although deGrom through his curveball less that 10% of the time in 2018, he was much more effective with it when he did. His strike percentage with the curveball jumped over 10% while the ball percentage and in play percentage dropped significantly as well. It is also quite interesting that although deGrom has increased his usage of his slider over the course of the data set, his “effectiveness” of the pitch has stayed relatively constant over the same time period.

Speeds on Strikeouts vs non-Strikeouts

To help with the clustering objective of this project I wanted to see if there were any patterns from some metrics on strike outs vs non-strikeouts so I first did density plots of pitch speed split up by pitch type and by years. The resulting plots are below The interesting trends I noticed in these density plots are the unimodal nature of a majority of the non-strikeout density plots and there is a higher proportion of bimodal density plots in the strikeout pitches which suggets that deGrom may be able to to vary his speeds on his pitches and will pick his spots to do so. This could also just be a result of much smaller sample size of strikeout pitches per year to non-strikeout pitches. However I still think these patterns can be exploited. Another really interesting component is the increased velocity deGrom experienced following the 2016 season. As I noted in my first quarter project, this is most likely due to an offseason surgery deGrom had to move a nerve in his elbow that was causing him some discomfort.

Spin rates on strikeout vs non-strikeout

Similarly I conducted the same type of analysis on spin rates but decided to use boxplots instead. It is interesting to note that spin rates of pitches that benefit from higher spin rates have improved spin rates when they are the strikeout pitch and the changeup which benefits from lower spin rate has a lower distribution when it is a strikeout pitch.

Spin Rate vs Release Speed

Many people in baseball I think would agree that spin rate and release speed of a pitch are two of the most important physical components of a pitch so I just wanted to see their relationship on a scatter plot over the seasons in the dataset.

The most interesting thing about this plot is the tightness of the clusters increases after each year. It appears that deGrom has just overall been a much more consistent pitcher.

Release Point

I am putting my most important finding from the first quarter project in the secondary findings for this project because I am unsure of its applicablity to this project. But my favorite visualization is shown below.

Like the spin rate vs speed visualization showed Jacob deGrom has become a much more consistent pitcher with his mechanics over the years and this is shown by the cluster of release points in 2018 being so tightly packed.

I look forward to finding more patterns in the data with my clustering models.